Like Two Porcupines Making Love—VERY CAREFULLY!

This was likely the best answer (and certainly the cleverest) response I read on a large cork board located in the registration area at this year’s ATA Annual Conference in Palm Springs. The board contained one main question to which attendees were encouraged to respond: “Do you use machine translation (MT), and how?” Here are some of the other comments (and some comments on the comments) attendees shared:

  • “Integrated in my computer-assisted translation tool. Pre-translation when nothing is found in translation memory. It needs very careful review.”
  • “To quickly get the gist of a paragraph. As a quick dictionary when I forget the target word in my native language.”
  • “For IT messages and informational messages.”
  • “Out of curiosity! The robots are coming whether we like it or not. We should know what MT is good or not good for.”
  • “To check genders (but we heard that’s not 100% accurate… )”
  • “For kicks.”
  • “To get a good laugh.”
    • ↑ That
  • “To enhance the skill of human linguists to reason over language and content and produce quality documentation of complex interactions.”
  • Irony, surely?
  • “To show how it does NOT work and can never replace the high quality professional work I do.”
    • APPLAUSE
  • “Integrated in my computer-assisted translation tool to assemble segments.”
    • Me +
    • Me three
  • “Never for any serious kind of translation work! For laughs—to get the gist of some Facebook posts and the likes only!”
  • “To show how algorithms can’t make the mental leaps that humans can! #vigoroustranslation #popthequestion”
  • “As a chainsaw, not a scalpel (i.e., only for a very rough job and not for anything delicate).”
  • “For the post-edited MT output of popular TV subtitles—at first I was incredulous that the domain would process well with MT—the results were surprisingly good!”
  • “As seldom as possible.”
  • “When I’m working on a translation in Adobe and need to understand the target or match it up with the source.”
  • “For triage—tells me at a glance what the topic of an article is. Will sometimes line-edit MT output if the nuance of the source isn’t important.”
  • “I don’t… yet!”
    • Me too!! Dinosaurs rule!!
    • Talk to me!
  • “Only when requested by the client.”
    • Then educate the client on MT, computer-assisted translation, and the difference between the two…and on human translation.
  • “No. Un-user-friendly—terrible human-machine interface.”
  • “Like using a lever to move heavy rock—you have to do the same amount of work, but you can distribute it more intelligently. You don’t have to exert the brute cognitive force of creating words ex nihilo.”
  • “I don’t—is that wrong of me?”
    • Not at all—very smart of you!

I thought it was very interesting to see some outside-the-box ideas in the comments (and the comments on the comments). I liked the idea of using MT as a quick aid to know where to place text in a document if you can’t read the language, or for gisting, or as a quick terminology resource. I also like the mention of different MT strategies for different projects and, again, I admire the porcupinian approach (you just learned a new word!).

These are all voices of professional translators, which implicitly makes them important and reflective of a professional reality, but here is my hope. I hope that next year and the year after, and the year after that, that there will be changes in the answers on similar cork boards. There will surely always be the “dinosaur”-like answers (I’m actually quoting a word from one of the comments rather than using a descriptive term myself). And there will also always be those answers that reflect the typical post-editing approach (editing one suggestion from one MT engine) as the chosen way to use MT. Percentages might shift a little toward the latter, but generally speaking those two voices are going to stay unchanged. But I long for more outside-the-box ideas.

I hope to see answers like using not one but several MT engines simultaneously (or several suggestions from one MT engine); or using auto-complete to just use fragments of MT suggestions; or using your termbase or a glossary to automatically correct or flag MT suggestions; or using your translation memory to give an MT match a reliability rating (or vice versa); or using voice recognition to work alongside and together with MT; or, or, or. (And here neither I nor anyone else knows what could possibly follow those open-ended “or’s.”)

The data generated by MT is a resource that can be used with varying levels of success along with other resources. In fact, it can be combined with other resources. One thing that sets translators apart from others is the creativity we use to produce our translations. Why not use that same creativity in the ways we approach our work? This work by definition is so diverse (after all, there’s nothing that can’t be translated), and that very diversity requires different approaches for every translator, and maybe even every project.

I want us to be creative in the way we use our resources. Think about the ways in which we use and maintain our termbases. While it’s true that, overall, we probably don’t use them enough (“too tedious and therefore too expensive to create and maintain”), we have very individual approaches to using them, even as to what kind of data we’re entering, how we’re using the data, and what our hopes are for it. Few professional translators would argue that terminology work is useless or has a nefarious purpose, even though it’s sometimes underused. It’s laughable to even think that way. When I talk to translators about terminology maintenance, I generally get only two responses. One is the slightly embarrassed admission that “I’m not using it nearly enough, even though I should,” and the other is “I love it and I can’t imagine how I could work without it.”

Now, there is no value in a terminology database per se. There is value in what we make of it. And while this is not a perfect parallel to MT engines, the “what-we-make-of-it” part of it is. Why aren’t we as adventurous in finding new and better ways to work with that resource as we are with other resources? (Which, as in the case of termbases, might mean that many don’t use it.)

Here are some thoughts that might help us. First, we need to understand that every situation is different. While it’s great that large companies train their MT engines and therefore have no problems with erratic terminology in the MT output, that’s not the case for the vast majority of translators. There are many translators working for TransPerfect and Lionbridge on the supplier side or Microsoft and GM on the buyer side, but many more do not. This means that the experience of one kind of translation doesn’t necessarily “translate” to another set of experiences. We need to really understand this when we talk to each other about MT, and any technology.

Then there are real differences in language combinations. While neural MT has leveled the playing field a little bit when it comes to the quality of languages that are syntactically very different, and were therefore a difficult nut to crack for previous kinds of MT, it has not solved the problem of low-resource languages, for instance. (I regret not asking for the language combination of those who responded to the MT question on the cork board at the conference.)

Then there is the kind of technology we use, both through MT and through the environment via which we access the MT (aka “translation environment tool”).

  • Is the MT adaptive or not?
  • Am I allowed to use certain kinds of MT?
  • Does my translation environment tool or my external plugin allow me to access one or several suggestions and, if so, how?
  • Do my terminology management, termbase, and MT suggestions “talk” to each other?
  • And so on and so forth

One of the reasons I’ve been very insistent on using the term “translation environment tool” is that “environment” comes with the concept that all kinds of features are available and can be used to the user’s liking. It provides wide open rooms that I can decorate as I like. I think it’s the job of the technology vendor to keep it that way and even broaden that approach. It’s our job to pick and choose our tools within those environments and become excellent at using them. And maybe even find a way to use a tool that is uniquely your own, because there might be no other colleague who uses it quite the way you do.


Jost Zetzsche is chair of ATA’s Translation and Interpreting Resources Committee. He is the author of Translation Matters, a collection of 81 essays about translators and translation technology. Contact: jzetzsche@internationalwriters.com.

This column has two goals: to inform the community about technological advances and encourage the use and appreciation of technology among translation professionals.

The ATA Chronicle © 2020 All rights reserved.